Model comparison
Gemma 3 4B vs GPT-5.2
GPT-5.2 is the stronger model overall, scoring 54.1 to 27.9 on the Noometry Index. Gemma 3 4B costs 96× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.
Last verified . 22 shared benchmarks.
GoogleSummary
- They share 22 benchmarks with published results for both. Gemma 3 4B scores higher in 0 categories and GPT-5.2 in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where GPT-5.2 leads 59.2 to 11.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 7.5% for Gemma 3 4B and 96.1% for GPT-5.2.
- Gemma 3 4B is cheaper at $0.04 / $0.08 per million input/output tokens, against $1.75 / $14 for GPT-5.2.
- GPT-5.2 accepts more context: 400K tokens versus 131K.
- Gemma 3 4B has downloadable open weights; the other is API-only.
Side by side
| Gemma 3 4B | GPT-5.2 | |
|---|---|---|
| Provider | OpenAI | |
| Noometry Index | 27.9 | 54.1 |
| Released | 2025-03-12 | 2025-12-11 |
| Weights | Open | Proprietary |
| Context window | 131K | 400K |
| Max output | 4K | 128K |
| Input $ / M tokens | $0.04 | $1.75 |
| Output $ / M tokens | $0.08 | $14 |
| Results tracked | 22 | 67 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.2 leads
Gemma 3 4B: 35.3 (#214), GPT-5.2: 52.2 (#32)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Coding | 1230 | 1447 |
| SWE-bench Verified | — | 73.8% |
| SWE-bench Verified (bash only) | — | 72.8% |
| LMArena WebDev | — | 1415 |
| SWE-bench Multilingual | — | 66.7% |
| GSO | — | 27.4% |
| WeirdML | — | 72.2% |
| ALE-Bench | — | 1,294 |
| AlgoTune | — | 2.05 |
Agentic & Tool Use GPT-5.2 leads
Gemma 3 4B: 20.7 (#134), GPT-5.2: 40.0 (#24)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 19.6% | 55.9% |
| Terminal-Bench | — | 64.9% |
| GDPval | — | 49.7% |
| Remote Labor Index | — | 2.5% |
| τ²-bench Airline | — | 83% |
| τ²-bench Banking | — | 32.2% |
| τ²-bench Retail | — | 81.6% |
| τ²-bench Telecom | — | 89.7% |
| DeepResearch Bench | — | 41.1% |
| LMArena Search | — | 1207 |
| METR Time Horizons | — | 75.3% |
| Vending-Bench 2 | — | 3,591 |
Reasoning GPT-5.2 leads
Gemma 3 4B: 12.7 (#330), GPT-5.2: 50.1 (#33)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| Kagi LLM Benchmark | 25.2% | 73.3% |
| Chess Puzzles | 0% | 49% |
| LMArena Hard Prompts | 1253 | 1445 |
| DTBench | 50.9% | 90.9% |
| LMCA | 2.8% | 43.9% |
| Epoch Capabilities Index | 116.02 | 153.45 |
| ARC-AGI-2 | — | 52.9% |
| SimpleBench | — | 45.8% |
| NYT Connections (extended) | — | 83.6% |
| ARC-AGI-1 | — | 86.2% |
| EnigmaEval | — | 10.4% |
| EBR-Bench | — | 23% |
| Mystery Game Puzzles | — | 23% |
| ForecastBench | — | 60.1 |
Math GPT-5.2 leads
Gemma 3 4B: 15.8 (#286), GPT-5.2: 59.8 (#32)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 7.5% | 96.1% |
| LMArena Math | 1239 | 1439 |
| FrontierMath (Tiers 1-3) | — | 67.4% |
| FrontierMath Tier 4 | — | 31.7% |
| MathArena Final-Answer Competitions | — | 72% |
| ProofBench | — | 15% |
| FrontierMath (Feb 2025 set) | — | 40.7% |
| FrontierMath Tier 4 (v1) | — | 18.8% |
Knowledge GPT-5.2 leads
Gemma 3 4B: 11.8 (#294), GPT-5.2: 59.2 (#29)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| GPQA Diamond | 23.2% | 91.4% |
| Vectara Hallucination Rate | 6.4% | 8.4% |
| LMArena Expert | 1223 | 1445 |
| Humanity's Last Exam | — | 27.8% |
| SimpleQA Verified | — | 37.1% |
Multimodal Not comparable
Gemma 3 4B: —, GPT-5.2: 50.9 (#7)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Vision | — | 1268 |
| VPCT | — | 84% |
| Furniture Assembly | — | 38.3% |
| LMArena Document | — | 1405 |
Multilingual GPT-5.2 leads
Gemma 3 4B: 42.7 (#189), GPT-5.2: 53.6 (#63)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Non-English | 1273 | 1425 |
| LMArena German | 1281 | 1448 |
| LMArena Russian | 1294 | 1440 |
| LMArena Chinese | — | 1460 |
| LMArena French | — | 1456 |
| LMArena Japanese | — | 1421 |
| LMArena Korean | — | 1390 |
| LMArena Spanish | — | 1433 |
Instruction Following GPT-5.2 leads
Gemma 3 4B: 65.4 (#219), GPT-5.2: 74.9 (#83)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Instruction Following | 1239 | 1417 |
Long Context GPT-5.2 leads
Gemma 3 4B: 38.6 (#189), GPT-5.2: 44.0 (#74)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Longer Query | 1274 | 1428 |
| CL-bench | — | 18.2% |
Writing & Preference GPT-5.2 leads
Gemma 3 4B: 42.2 (#234), GPT-5.2: 67.0 (#30)
| Benchmark | Gemma 3 4B | GPT-5.2 |
|---|---|---|
| LMArena Text | 1291 | 1439 |
| LMArena Creative Writing | 1270 | 1402 |
| EQ-Bench Creative Writing | 1068 | 1703 |
| LMArena Multi-Turn | 1255 | 1458 |
Frequently asked questions
Is Gemma 3 4B better than GPT-5.2?
GPT-5.2 is the stronger model overall, scoring 54.1 to 27.9 on the Noometry Index. Gemma 3 4B costs 96× less per token, which makes it the better buy when GPT-5.2's lead doesn't matter for your workload.
Which is cheaper, Gemma 3 4B or GPT-5.2?
Gemma 3 4B is cheaper. It lists at $0.04 per million input tokens and $0.08 per million output tokens; GPT-5.2 lists at $1.75 and $14.
Is Gemma 3 4B or GPT-5.2 better for coding?
GPT-5.2 scores higher on coding benchmarks: 52.2 versus 35.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5.2 does, with 400K tokens against 131K.
How many benchmarks do Gemma 3 4B and GPT-5.2 share?
22 benchmarks have published results for both models. Gemma 3 4B has 22 scored results on Noometry and GPT-5.2 has 67.